Orange Business: We Drink Our Own Orange Juice

When Orange Business CEO Christel Heydemann says the company runs its own operations on the same Live Intelligence platform it sells to enterprise customers — "We drink our own orange juice," she put it in a LinkedIn post — she is trying to answer the question every European company is asking right now: who do you trust to run your AI agents?

The answer Orange Business is betting on, announced at its 2026 summit in Paris in March, is Live Intelligence Studio: a platform built on LangChain's open-source framework stack (LangChain, LangGraph, and LangSmith for observability) that lets enterprise teams design, deploy, and govern autonomous AI agents inside Orange's European sovereign infrastructure. Orange Business press release Harrison Chase, LangChain's CEO, confirmed the partnership directly: "We are proud to work with Orange Business, combining our open source frameworks (LangChain and LangGraph) with LangSmith's agent observability and evaluation capabilities, all integrated with the trusted Live Intelligence platform." Harrison Chase quote from orange-business.com That quote — from the framework vendor, not from Orange's PR shop — is the closest thing to external validation this announcement has, because Orange has not named a single enterprise customer for Live Intelligence Studio specifically. The platform more broadly serves over 100 business customers and 100,000 users, but those numbers include the existing Live Intelligence product lineup, not the new Studio tier.

None of which makes this a bad story. It makes it the right kind of story, which is an orchestration layer story — not an AI model story.

The 80 million conversations are the real evidence

Here is the number that should focus attention: Orange Business managed 80 million AI conversations in 2025. That is not a pilot. That is not a waitlist. That is volume running through a voice infrastructure product today. Combined with the fact that 80 percent of customer service agents still say voice is their top-used channel on a daily basis — per research from Cavell Enable 2025 — it suggests that the "agentic telephony" layer, where AI manages call context and autonomously navigates the call journey, is the concrete product in this announcement, not the agent builder.

Usman Javaid, Chief Product and Marketing Officer at Orange Business, described the shift this way: "We are moving from AI chatbots to a new era where AI agents can autonomously perform complex tasks. This marks a fundamental shift for enterprises: AI is no longer just a productivity tool, but a new layer of intelligent automation that augments teams and executes complete workflows." Orange Group press release That is the pitch. The 80 million conversations is the evidence it is not theoretical.

What Orange is actually selling

The architecture is worth understanding precisely because it is not a model play. Orange is not claiming to have built a foundation model. It is claiming to have built the sovereign wrapper layer — the infrastructure that sits between European enterprises and the hyperscalers, handling agent orchestration, observability, cost control, and quality monitoring while the underlying LLMs run elsewhere. LangGraph, the stateful workflow component of LangChain, is the critical piece here: it adds graph structures for complex multi-agent workflows, loops, retries, and persistence, which is what separates a production-grade agent deployment from a prototype chatbot. Orange is positioning itself as the compliance and sovereignty wrapper around a stack whose core components — LangChain, the LLMs — come from elsewhere.

The deeper play is European data residency. Orange Business serves over 7,000 enterprise customers and 100,000 customer locations globally across its voice and customer experience business. Orange Group as a whole generated 40.4 billion euros in revenues as of end 2025 and connects 340 million customers across 26 countries. That footprint is the product. A European manufacturer that wants agentic AI infrastructure without sending its operational data to a US hyperscaler has fewer options than the announcement implies, but Orange is one of them, and it has the telco relationships to make the voice layer work as a single contract rather than a patchwork integration.

What is actually live versus what is coming

The announcement also includes deepfake detection integrated at the network layer — covering fake audio, images, video, and documents — via partners Sensity and Reality Defender. Orange Business press release Network-layer detection intercepts media at the telecom level before it reaches the endpoint, which is architecturally different from on-device scanning — it covers all calls simultaneously but raises different privacy considerations. The problem is the announcement's own language: "Moving forward," Orange's press release says, "the platform will introduce advanced protections, such as AI agent security and deepfake detection." Will. Not has. This is a future capability. The wire summary implied it was live infrastructure. It is not. The architectural ambition — network-layer detection at telecom scale — is interesting. The evidence it exists today is not in this announcement.

The agentic telephony integrations with Microsoft 365 Copilot agents in Teams (including the Interpreter agent) and planned extension to Cisco Webex are live-adjacent in a more credible way: those partnerships involve existing platforms Orange is extending rather than capabilities it is building from scratch. Orange Business press release

The competitive lane is getting crowded

Orange is not alone in this positioning. Deutsche Telekom and NVIDIA launched the Industrial AI Cloud in November 2025, going live in early 2026, targeting sovereign AI infrastructure for European enterprises. NVIDIA's page on sovereign AI infrastructure lists not just Orange but also Fastweb, Swisscom, Telefónica, and Telenor as telcos building sovereign AI factories and edge infrastructure across Europe. The European telco-as-sovereign-AI-layer play is a real trend — Orange is one of several companies making the same bet simultaneously. The differentiator will not be the positioning, which is shared, but whether the deployment volumes follow.

The Heydemann trust signal

Christel Heydemann's "drinks her own orange juice" comment is the kind of line that sounds like marketing until you think about it for thirty seconds. In enterprise AI infrastructure, the reason internal use matters as a trust signal is that it means Orange's own operations team has already hit whatever compliance, data residency, and performance requirements its enterprise customers have — and survived. If the platform had failed internally, Heydemann would not be citing it. RCR Wireless Which one is the real answer is the most interesting open question this announcement leaves.

What Orange Business is building — a sovereign orchestration layer between European enterprises and the AI model providers — is a coherent product strategy in a market where that need is real. The 80 million conversations are evidence that the underlying infrastructure works. The LangChain partnership gives it a credible framework stack. The competitive lane is crowded. The deepfake detection is not live yet. The enterprise customer names for Live Intelligence Studio specifically are absent. But the architecture is coherent and the volume is real. That is more than most AI infrastructure announcements ship with.